Table 3 Features used to estimate phase confidence. The top 10 features, ranked by their importance (GINI importance, as calculated by scikit-learn [10] for random forests)

Proportion of database individuals with IBD segments assigned to both sides of the family, both on the largest supercluster

Proportion of IBD segments that are partially assigned to one parental side and partially to the other

Number of close family members that do not share DNA with all other close family

(log) number of database individuals with shared DNA

(log) number of IBD segments assigned to the largest supercluster

Number of missing edges in close family network (pairs of close family that do not share IBD with each other)

Proportion of the genome overlapped by at least one IBD segment

Ratio of the number of database individuals with shared DNA on one IBD segment to the number with shared DNA on multiple IBD segments

Proportion of the genome overlapped by at least two IBD segments

Proportion of IBD segments assigned to the largest supercluster